Live attack telemetry

Every attack one small internet-facing server (testingpenguins.com) has caught and blocked, from a security data pipeline I built and run. Nothing here is simulated.

Loading…

Where the attacks come from

Each dot is an attacker location; size = number of attacks, color = what most of them tried.

Attacks per day

Local detections by type. SSH brute force disappears after August: SSH is no longer reachable from the internet, so there is nothing left to attack.

What attackers try

Top detection scenarios, by number of attacks.

Who spotted them first

This server runs CrowdSec, which also shares a community blocklist. For each attacker: did this server catch it before the community listed it?

Top countries

By distinct attacker IPs. Location = where the IP is registered, not who is behind it.

Top networks

Attackers mostly rent cloud servers; scanners like Censys show up too.

Repeat offenders

IPs banned, released when the ban expired, and banned again for coming back. IPs are partly masked.

How this works

CrowdSecreads web & system logs, detects and bans
→
Apache Airflowhourly ETL (extract, transform, load)
→
PostgreSQLalerts, events, ban timeline
→
This pagestatic JSON, rebuilt hourly

An Airflow DAG (directed acyclic graph) pulls detections and the full ban list from CrowdSec every hour, models them in PostgreSQL (including a ban timeline that stores each ban as a time range instead of hourly copies, about 140× smaller), and publishes this page's data. The page itself is plain static files: no database or server code is reachable from the internet.

Data notes

Locations come from IP geolocation and are approximate (often city or country level); IPs with only country-level data are drawn at the country's center. "Attacks" are CrowdSec detections (one IP can trigger several). Repeat-offender IPs are masked to their first two octets. Times are US Eastern. The arcs in the replay end at a generic point for this server, not its real location.